Joint IDs Embedding and its Applications in E-commerce
share
E-commerce has become an important part of our daily lives and there are great challenges due to its dynamic and complex business environment. Many machine intelligence techniques are developed to overcome these challenges. One of the essential elements in those techniques is the representation of data, especially for ID-type data, e.g. item ID, product ID, store ID, brand ID, category ID etc. The classical one-hot encoding suffers sparsity problems due to its high dimension. Moreover, it cannot reflect the relationships among IDs, either homogeneous or heterogeneous ones. In this paper, we propose a novel hierarchical embedding model to jointly learn low-dimensional representations for different types of IDs from the implicit feedback of users. Our approach incorporates the structural information among IDs and embeds all types of IDs into a semantic space. The low-dimensional representations can be effectively extended to many applications including recommendation and forecast etc. We evaluate our approach in several scenarios of "Hema App" and the experimental results validate the effectiveness of our approach.
READ FULL TEXT
